Agentic · tools

OSWorld-Verified leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on OSWorld-Verified is 86.1% by Qwen3.8 Max. 143 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude Fable 5.1	100.0%	estimated ± 1.9 pp, low confidence
2	Claude Opus 5.5	100.0%	estimated ± 3.0 pp, medium confidence
3	Gemini 4 Argon	100.0%	estimated ± 3.0 pp, medium confidence
4	GPT-6 Sol	100.0%	estimated ± 3.0 pp, medium confidence
5	GPT-6 Astra	99.7%	estimated ± 1.9 pp, low confidence
6	Claude Sonnet 5.5	90.7%	estimated ± 8.7 pp, low confidence
7	Grok 4.7	87.5%	estimated ± 8.7 pp, low confidence
8	Claude Opus 5	86.4%	estimated ± 1.9 pp, low confidence
9	Qwen3.8 Max Preview	86.3%	estimated ± 8.7 pp, low confidence
10	Qwen3.8 Max	86.1%	measured
11	GPT-5.5 Pro	85.3%	estimated ± 5.4 pp, low confidence
12	Claude Fable 5	85.0%	measured
13	Claude Mythos 5	85.0%	measured
14	GPT-5.4 Pro	84.3%	estimated ± 5.4 pp, low confidence
15	Qwen3.8-27B	84.3%	measured
16	GPT-6.1 Sol	83.8%	estimated ± 8.7 pp, low confidence
17	Claude Opus 4.8	83.4%	measured
18	Gemini 3.6 Flash	83.0%	measured
19	Qwen3.8-Flash-Next	82.8%	estimated ± 3.0 pp, high confidence
20	Holo3-35B-A3B	82.6%	measured
21	GPT-5.6 Sol	82.2%	estimated ± 1.9 pp, medium confidence
22	MiMo-V2.6-Pro	82.0%	measured
23	Gemini 3.8 Flash	81.3%	estimated ± 1.9 pp, medium confidence
24	Muse Spark 1.2	81.2%	estimated ± 8.7 pp, low confidence
25	Claude Sonnet 5	81.2%	measured
26	Ornith-1.5-397B	80.9%	estimated ± 5.4 pp, medium confidence
27	MiMo-V2.6-Flash	80.8%	measured
28	Muse Spark 1.1	80.8%	measured
29	DeepSeek V4.1 Flash	80.7%	estimated ± 4.7 pp, high confidence
30	Ling 3.1 Flash	80.7%	estimated ± 4.7 pp, high confidence
31	Fugu Cyber	80.5%	estimated ± 4.7 pp, high confidence
32	Atria Dawn Preview	80.4%	estimated ± 4.7 pp, high confidence
33	Gemini 3.8 Flash Cyber	80.3%	estimated ± 4.7 pp, high confidence
34	GPT-6 Luna	80.2%	estimated ± 8.7 pp, low confidence
35	Step 5 Preview	80.0%	estimated ± 4.7 pp, high confidence
36	GLM-5.3	80.0%	estimated ± 4.7 pp, high confidence
37	Mistral Large 4	79.8%	estimated ± 8.7 pp, low confidence
38	DeepSeek V4 Pro 0813	79.7%	estimated ± 4.7 pp, high confidence
39	Gemini 3.5 Flash Cyber	79.7%	estimated ± 4.7 pp, high confidence
40	Claude Mythos Preview	79.7%	estimated ± 4.7 pp, high confidence
41	Apodex 1.1	79.3%	estimated ± 6.7 pp, low confidence
42	Muse Spark 1.3	79.3%	estimated ± 1.9 pp, medium confidence
43	Grok 4.5	78.9%	estimated ± 8.7 pp, low confidence
44	Holo3-122B-A10B	78.9%	measured
45	Kimi K3	78.8%	estimated ± 1.9 pp, medium confidence
46	GPT-5.5	78.7%	measured
47	Hy4 preview	78.6%	estimated ± 4.7 pp, high confidence
48	GLM-5.2	78.5%	estimated ± 8.7 pp, low confidence
49	Gemini 3.5 Flash	78.4%	measured
50	DeepSeek V4 Flash 0731	78.2%	estimated ± 4.7 pp, high confidence
51	Gemini 3.7 Flash	78.0%	estimated ± 1.9 pp, medium confidence
52	GPT-5.6 Terra	78.0%	estimated ± 1.9 pp, medium confidence
53	Claude Opus 4.7 (Adaptive)	78.0%	measured
54	UI-Mate-27B	77.0%	measured
55	Grok 4.6	76.8%	estimated ± 1.9 pp, medium confidence
56	dots3-note Preview	76.6%	estimated ± 5.4 pp, medium confidence
57	GLM-5.1	76.0%	estimated ± 4.7 pp, high confidence
58	GPT-5.4	75.0%	measured
59	Claude Opus 4.7	74.2%	estimated ± 1.9 pp, medium confidence
60	GPT-5.6 Luna	74.2%	estimated ± 1.9 pp, medium confidence
61	Qwen3.7 Max	74.0%	estimated ± 5.9 pp, medium confidence
62	Gemini 3.5 Flash-Lite	74.0%	measured
63	Apodex 1.1 Mini	73.8%	estimated ± 8.7 pp, low confidence
64	Quasar 438B	73.5%	estimated ± 8.7 pp, low confidence
65	Qwen3.7 Plus	73.3%	measured
66	Kimi K2.6	73.1%	measured
67	Gemini 3 Pro	73.0%	estimated ± 5.9 pp, medium confidence
68	Ling 3.0 Flash VL	73.0%	estimated ± 8.7 pp, low confidence
69	Claude Opus 4.6	72.7%	measured
70	MiMo-V2.5-Pro	72.5%	estimated ± 5.9 pp, medium confidence
71	Claude Sonnet 4.6	72.1%	measured
72	GPT-5.4 mini	72.1%	measured
73	Hy3	70.4%	estimated ± 8.7 pp, low confidence
74	MiniMax M3	70.1%	measured
75	Kimi K2.7 Code	69.7%	estimated ± 8.7 pp, low confidence
76	Inkling-Small	69.1%	estimated ± 5.4 pp, medium confidence
77	Beam	69.1%	estimated ± 5.4 pp, medium confidence
78	GLM-5.3-Flash	69.0%	estimated ± 6.2 pp, low confidence
79	Inkling	68.7%	estimated ± 5.4 pp, medium confidence
80	A.X K2	67.7%	estimated ± 8.7 pp, low confidence
81	Ling 3.0 Flash FP8	67.3%	estimated ± 8.7 pp, low confidence
82	Step 3.7 Flash	67.1%	estimated ± 5.4 pp, medium confidence
83	Agents-A1	66.8%	estimated ± 5.4 pp, medium confidence
84	Gemini 3.1 Pro	66.8%	estimated ± 5.9 pp, medium confidence
85	Gemini 3 Flash	66.5%	estimated ± 5.9 pp, medium confidence
86	Claude Opus 4.5	66.3%	measured
87	UI-Mate-9B	66.2%	measured
88	Solar Open 2	66.0%	estimated ± 10.9 pp, low confidence
89	Muse Glimmer 30B	65.9%	measured
90	Muse Spark	65.8%	estimated ± 4.7 pp, medium confidence
91	GLM-5	65.6%	estimated ± 4.7 pp, medium confidence
92	Qwen3.6-27B	64.8%	estimated ± 5.9 pp, medium confidence
93	GPT-5.3 Codex	64.7%	measured
94	Ling 3.0 Flash	63.0%	estimated ± 5.4 pp, medium confidence
95	GPT-5.2-Codex	61.9%	estimated ± 5.9 pp, medium confidence
96	Claude Sonnet 4.5	61.4%	measured
97	MiniCPM5-2B	61.1%	estimated ± 8.7 pp, low confidence
98	Qwen3.6 Plus	60.9%	estimated ± 5.9 pp, medium confidence
99	LLaDA2.2-flash	60.2%	estimated ± 10.9 pp, low confidence
100	GPT-5.1-Codex	60.1%	estimated ± 5.9 pp, medium confidence
101	Grok Build 0.1	59.7%	estimated ± 5.9 pp, medium confidence
102	MiMo-V2-Flash	59.0%	estimated ± 8.7 pp, low confidence
103	Grok 4.1 Fast	58.3%	estimated ± 5.9 pp, medium confidence
104	Ornith-1.5-35B-A3B	58.3%	estimated ± 5.4 pp, medium confidence
105	MiMo-V2.5	58.0%	estimated ± 5.9 pp, medium confidence
106	Qwen3.5-122B-A10B	58.0%	measured
107	Granite 4.2 30B	57.6%	estimated ± 8.7 pp, low confidence
108	Agents-A1-4B	57.6%	estimated ± 5.4 pp, medium confidence
109	Gemma 4 26B A4B	57.3%	estimated ± 8.7 pp, low confidence
110	Ling 3.0 Tiny	57.2%	estimated ± 8.7 pp, low confidence
111	Qwen3.5-27B	56.2%	measured
112	Grok 4.3	56.2%	estimated ± 5.9 pp, medium confidence
113	Qwen3 Max	56.1%	estimated ± 5.9 pp, medium confidence
114	Command A+	56.0%	estimated ± 8.7 pp, low confidence
115	Celeris-1	55.6%	estimated ± 8.7 pp, low confidence
116	DeepSeek V3	55.6%	estimated ± 8.7 pp, low confidence
117	DeepSeek V3 0324	55.6%	estimated ± 8.7 pp, low confidence
118	Gemma 3 27B	55.6%	estimated ± 8.7 pp, low confidence
119	Gemma 4 12B	55.6%	estimated ± 8.7 pp, low confidence
120	Gemma 4 E2B	55.6%	estimated ± 8.7 pp, low confidence
121	Gemma 4 E4B	55.6%	estimated ± 8.7 pp, low confidence
122	GPT-4.1 mini	55.6%	estimated ± 8.7 pp, low confidence
123	GPT-4.1 nano	55.6%	estimated ± 8.7 pp, low confidence
124	GPT-4o	55.6%	estimated ± 8.7 pp, low confidence
125	GPT-4o mini	55.6%	estimated ± 8.7 pp, low confidence
126	GPT-OSS 20B	55.6%	estimated ± 8.7 pp, low confidence
127	Granite 4.2 3B	55.6%	estimated ± 8.7 pp, low confidence
128	Granite 4.2 8B	55.6%	estimated ± 8.7 pp, low confidence
129	K-Exaone	55.6%	estimated ± 8.7 pp, low confidence
130	LFM2.5-2.6B	55.6%	estimated ± 8.7 pp, low confidence
131	Ling 2.6 Flash	55.6%	estimated ± 8.7 pp, low confidence
132	Llama 4 Maverick	55.6%	estimated ± 8.7 pp, low confidence
133	Llama 4 Scout	55.6%	estimated ± 8.7 pp, low confidence
134	Mistral Large 3	55.6%	estimated ± 8.7 pp, low confidence
135	Mistral Small 4	55.6%	estimated ± 8.7 pp, low confidence
136	Mistral Small 4 (Reasoning)	55.6%	estimated ± 8.7 pp, low confidence
137	Nemotron 3 Nano 30B	55.6%	estimated ± 8.7 pp, low confidence
138	Nemotron 3 Nano Omni 30B A3B	55.6%	estimated ± 8.7 pp, low confidence
139	Nemotron 3 Super 100B	55.6%	estimated ± 8.7 pp, low confidence
140	North Mini Code	55.6%	estimated ± 8.7 pp, low confidence
141	Solar Pro 3	55.6%	estimated ± 8.7 pp, low confidence
142	Trinity-Large-Preview	55.6%	estimated ± 8.7 pp, low confidence
143	Ultravox v0.6 Llama 3.3 70B	55.6%	estimated ± 8.7 pp, low confidence
144	Qwen3.6-35B-A3B	55.5%	estimated ± 5.9 pp, medium confidence
145	Claude 4.1 Opus	55.4%	estimated ± 8.7 pp, low confidence
146	Grok 4	55.4%	estimated ± 5.9 pp, medium confidence
147	Gemini 2.5 Pro	55.2%	estimated ± 5.9 pp, medium confidence
148	GPT-5.1	54.9%	estimated ± 5.9 pp, medium confidence
149	MiniMax M2.7	54.5%	estimated ± 5.9 pp, medium confidence
150	Qwen3.5-35B-A3B	54.5%	measured
151	Claude 4 Sonnet	54.2%	estimated ± 5.9 pp, medium confidence
152	Mistral Medium 3.5 128B	54.0%	estimated ± 5.9 pp, medium confidence
153	Gemini 3.1 Flash-Lite	53.8%	estimated ± 5.9 pp, medium confidence
154	Grok 4.20	53.8%	estimated ± 5.9 pp, medium confidence
155	Qwen3.5 397B	53.8%	estimated ± 5.4 pp, medium confidence
156	Hy3 Preview	53.4%	estimated ± 5.9 pp, medium confidence
157	MiMo-V2-Pro	53.3%	estimated ± 5.9 pp, medium confidence
158	Gemma 4 31B	53.0%	estimated ± 5.9 pp, medium confidence
159	Kimi K2.5	52.8%	estimated ± 5.4 pp, low confidence
160	Kimi K2.5 (Reasoning)	52.8%	estimated ± 5.4 pp, low confidence
161	Trinity-Large-Thinking	52.5%	estimated ± 5.9 pp, medium confidence
162	GLM-5V-Turbo	52.2%	estimated ± 5.9 pp, medium confidence
163	GPT-OSS 120B	52.1%	estimated ± 5.9 pp, medium confidence
164	DeepSeek V3.2	52.1%	estimated ± 5.9 pp, medium confidence
165	GPT-4.1	51.8%	estimated ± 5.9 pp, low confidence
166	Ornith-1.5-9B	50.5%	estimated ± 5.4 pp, low confidence
167	Qwen3.5 Plus	50.5%	estimated ± 8.7 pp, low confidence
168	GLM-4.7	48.8%	estimated ± 5.4 pp, low confidence
169	Solar Pro 4	47.9%	estimated ± 5.4 pp, low confidence
170	LongCat-Flash-Lite-Sparse	47.8%	estimated ± 5.4 pp, low confidence
171	GPT-5.2	47.3%	measured
172	Nemotron 3 Ultra	47.0%	estimated ± 5.4 pp, low confidence
173	Claude Haiku 4.5	46.3%	estimated ± 8.7 pp, low confidence
174	Mercury 2.5	46.2%	estimated ± 6.7 pp, low confidence
175	Nemotron 3.5 Lightning 30B A3B NVFP4	46.2%	estimated ± 5.4 pp, low confidence
176	GPT-5.4 nano	39.0%	measured
177	GPT-5 (high)	30.2%	estimated ± 8.7 pp, low confidence

OSWorld-Verified leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General